Ensemble Deep Learning Framework for Multiclass Classification of Femur and Pelvic Fractures

Abstract

The fractures of the femur and pelvis are life-threatening orthopedic conditions that are common among older individuals, causing severe complications and reducing mobility. This paper introduces a deep learning method for automati cally recognizing pelvic and femur fractures in X-ray images. Our approach relies on ensemble deep learn ing models such as convolutional neural network (CNN) architectures, including ResNet50, Incep tionV3, ResNet101, EfficientNetB0, EfficientNetV2, MobileNet and Xception, which classify fractures into five possible categories: non-displaced, incom plete non-displaced, complete non-displaced, par tially displaced, and fully displaced fractures. Data, comprising around 1000 X-ray images from vari ous hospitals, were pre-processed and augmented to strengthen the model. The ResNet50 model achieved the highest classification accuracy at 80% on the test set and was identified as the best-performing model for distinguishing fracture types. The framework combines modern feature engineering and ensemble learning models to enable early and accurate diag nosis, leading to improved clinical outcomes in treat ing femur and pelvic fractures, reducing diagnostic errors, and significantly enhancing the diagnostic process.